Security patches

.github/workflows/job-arrow-extension.yml

@@ -45,7 +45,7 @@ jobs:
         uses: ./.github/actions/rust-extension-toolchain
 
       - name: Cache cargo registry and build
-        uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
+        uses: Swatinem/rust-cache@d1527c8f67ef9af357290475d4af4b0fab8f12e0 # v2.9.3
         with:
           # ext-php-rs bindings are generated against the PHP headers, so builds are per PHP version
           key: php${{ matrix.php }}

.github/workflows/job-flow-php-extension.yml

@@ -45,7 +45,7 @@ jobs:
         uses: ./.github/actions/rust-extension-toolchain
 
       - name: Cache cargo registry and build
-        uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2
+        uses: Swatinem/rust-cache@d1527c8f67ef9af357290475d4af4b0fab8f12e0 # v2.9.3
         with:
           # ext-php-rs bindings are generated against the PHP headers, so builds are per PHP version
           key: php${{ matrix.php }}

.gitignore

@@ -7,4 +7,5 @@ phpunit.xml
 phpbench.json
 benchmarks/datasets/
 benchmarks/var/
-.env
+.env
+/.github/tools/azurite/node_modules/

benchmarks/materialise.php

@@ -7,9 +7,6 @@
 
 require __DIR__ . '/../bootstrap.php';
 
-// Datasets::reset() (benchmarks/bootstrap.php) wipes benchmarks/var and refuses any path that does
-// not contain /benchmarks/var. Fixtures live in benchmarks/datasets, so nothing here interacts with it.
-
 $rows = BenchmarkRows::count();
 $windowRows = BenchmarkRows::windowCount();
 

benchmarks/src/BenchmarkRows.php

@@ -7,11 +7,7 @@
 use function getenv;
 
 /**
- * The row counts every bench provider and materialise.php resolve through. They must agree: a fixture
- * set materialised for a different count leaves the first timed run generating one.
- *
- * phpbench's own --parameters cannot carry this - it replaces a subject's whole parameter set rather
- * than merging into it, so any subject with a second provider would lose that axis.
+ * Providers and materialise.php must agree; phpbench --parameters replaces rather than merges a parameter set.
  */
 final class BenchmarkRows
 {

benchmarks/src/Datasets/DigestCache.php

@@ -5,8 +5,7 @@
 namespace Flow\Benchmarks\Datasets;
 
 /**
- * One source-tree walk per tree per process. Warmed from a #[Bench\BeforeMethods] method so no
- * timed window ever pays for it.
+ * Warm from #[BeforeMethods] so no timed window pays for the tree walk.
  */
 final class DigestCache
 {

benchmarks/src/Datasets/FixtureFormat.php

@@ -9,14 +9,6 @@
  */
 enum FixtureFormat: string
 {
-    /**
-     * Every derived fixture is from_parquet(root) -> to_X(), and the root itself is
-     * FakeRandomOrdersExtractor -> to_parquet, so these six trees are in the byte-producing
-     * chain of every fixture regardless of its format. Floe lives inside src/core/etl/src.
-     *
-     * benchmarks/src/Datasets is in the chain because TextDataset and SellersDataset build their
-     * rows in PHP rather than delegating to an adapter, so their bytes change with this tree.
-     */
     private const CHAIN = [
         'src/core/etl/src',
         'src/core/etl/tests/Flow/ETL/Tests/Double/FakeRandomOrdersExtractor.php',

benchmarks/src/Datasets/FixturePath.php

@@ -9,8 +9,7 @@
 use function glob;
 
 /**
- * The single place a fixture path is built. The fingerprint lives in the filename rather than in a
- * sidecar so it can never drift from the bytes it describes.
+ * Fingerprint lives in the filename so it cannot drift from the bytes.
  */
 final readonly class FixturePath
 {

benchmarks/src/Partitioning/PartitionCardinality.php

@@ -10,8 +10,7 @@
 use function Flow\ETL\DSL\ref;
 
 /**
- * The orders fixture spreads rows over 5 sellers and a full year of created_at, so these two keys
- * give ~5 and ~365 partitions. There is no synthetic fixture behind a third point and none is invented.
+ * Orders fixture gives ~5 (seller) and ~365 (created_at day) partitions; no third fixture exists.
  */
 enum PartitionCardinality: string
 {

benchmarks/src/Partitioning/PartitionedTree.php

@@ -18,12 +18,7 @@
 use function substr;
 
 /**
- * A partitioned tree is a derived fixture, so it lives in the fixture cache under a fingerprinted
- * name rather than in benchmarks/var - which Datasets::reset() wipes at every process start, making
- * every phpbench iteration rebuild the whole tree before it could measure a read.
- *
- * The read path must be a glob. from_csv() over a bare partitioned directory silently reads 0 rows,
- * and a following partition filter() then fails at bind: "Schema definition for entry ... not found."
+ * Lives in fixture cache (benchmarks/var is wiped per process); read path must be a glob or from_csv reads 0 rows.
  */
 final class PartitionedTree
 {

benchmarks/src/Pipeline/CanonicalPipeline.php

@@ -15,17 +15,7 @@
 use function uniqid;
 
 /**
- * The one pipeline shape every source is measured through, so a source's number differs only by its
- * read leg. Every subject that claims to be comparable with the others must run through here.
- *
- * withColumn is the only verb that adds a column, so it forces schema widening at plan-bind time.
- * filter is deliberately non-selective - created_at is never null - so the row count reaching the loader
- * cannot move for a second reason. select re-converges every source onto the same five columns, so the
- * loader does identical work whatever was read. Floe stores typed values, adding no stringification the
- * source had not already paid.
- *
- * The output path is unique per call because a fixed path re-run across iterations and revolutions would
- * measure overwrite behaviour.
+ * Every comparable subject runs through here so sources differ only by their read leg.
  */
 final readonly class CanonicalPipeline
 {

benchmarks/src/Pipeline/LimitPushdownScenario.php

@@ -9,9 +9,7 @@
 use function Flow\ETL\DSL\data_frame;
 
 /**
- * A null limit runs the same pipeline with no limit() in the plan, so the pair prices what a
- * limit left in the plan costs on a local source. The remote half of the question needs
- * push-down, which this engine does not have yet.
+ * Null limit = same plan without limit(); the pair prices a limit on a local source (no push-down yet).
  */
 final readonly class LimitPushdownScenario
 {

benchmarks/src/Pipeline/NodePipelineScenario.php

@@ -20,11 +20,7 @@
 use function uniqid;
 
 /**
- * XML and Text do not infer and cannot carry the orders schema, so they form their own family: the
- * same four verbs in the same order as the canonical shape, over one column.
- *
- * These two numbers are comparable to each other and to nothing else. The loader is to_text because a
- * one-column string frame in Floe would measure Floe's footer rather than the pipeline.
+ * XML/Text cannot carry the orders schema: comparable only to each other; to_text avoids measuring Floe's footer.
  */
 final readonly class NodePipelineScenario
 {

benchmarks/src/Pipeline/OrdersSchema.php

@@ -22,16 +22,7 @@
 use function Flow\Types\DSL\type_structure;
 
 /**
- * The WEAK schema — for each source, exactly what that source's own inference produces today.
- * Declaring it prices the sampling pass and nothing else; a stronger schema would additionally price
- * type strength, which is a different question. Frozen as literals rather than computed at bench
- * time, so inference drift fails loudly instead of quietly measuring something new. Arms that
- * coincide today are still independent facts — do not fold them into a shared helper.
- *
- * Every column of every source infers nullable, so there is no separate nullability rule.
- * SourceExtractorTest asserts every Source arm still equals what that source infers. The two
- * ServiceSource arms have no such guard: proving them needs a live database, so it cannot live in
- * the unit suite.
+ * Frozen literals of each source's own inference (SourceExtractorTest guards them; ServiceSource arms unguarded).
  */
 final readonly class OrdersSchema
 {

benchmarks/src/Pipeline/PlanDepthScenario.php

@@ -12,10 +12,7 @@
 use function Flow\ETL\DSL\ref;
 
 /**
- * schema() answers from the plan without executing it, so this prices Planner::plan() alone (optimize, translate,
- * bind, split) rather than $depth withColumn calls per row.
- *
- * One call plans PLANS identical plans, because a single plan is far below timer resolution.
+ * schema() plans without executing, so this prices Planner::plan() alone.
  */
 final readonly class PlanDepthScenario
 {